The New Zealand government has set a target of increasing the number of electric vehicles (EVs) in New Zealand to 64,000 by 2021 (Transpower New Zealand 2017). High penetration of EVs would cause EV recharging to contribute a substantial portion of total electricity load. A report prepared for lines companies Orion, Powerco and Unison by Concept Consulting Group entitled “Driving change - Issues and options to maximise the opportunities from large-scale electric vehicle uptake in New Zealand” predicts that if all current light private vehicles were electric, annual residential electricity consumption would increase by approximately 30%, whereas if all vehicles including trucks were electric, this would increase the total electricity consumption of New Zealand by approximately 41% (Concept Consulting 2018).
New Zealand’s total electricity demand varies throughout the day, with weekdays in particular having two distinct “peaks”; one in the morning, and one in the evening (Transpower New Zealand 2015). Providing the electricity to meet these demand peaks is a costly and inefficient process (Khan, Jack, and Stephenson 2018). Concurrent electric vehicle charging, especially in the early evening when many motorists return home (Speidel and Bräunl 2014,Langbroek, Franklin, and Susilo (2017)), would have the potential to negatively impact the operation of the grid through drastically increasing peak loads (Azadfar, Sreeram, and Harries 2015,Langbroek, Franklin, and Susilo (2017)), leading to an increased cost of electricity due to the requirement of expensive upgrades to the electricity grid (Stephenson et al. 2017).
The Concept Consulting report considers different methods of EV charging in its models. The assumption that most drivers would begin charging immediately after returning home is referred to as “passive” charging, while charging that is programmed (either by the driver or by an external entity) to occur during off-peak periods is referred to as “smart”. The modelling undertaken in the Concept Consulting report suggests that under a scenario whereby 57% of the current private vehicle fleet were EVs (corresponding to one EV per household), passive charging would cause an increase of peak electricity demand of approximately 3,000MW, whereas if all were charged in a “smart” fashion, there would be no increase in peak demand.
This report extends the work done by Concept Consulting, but utilises actual data collected from electric vehicles, as opposed to using models based on the current New Zealand transport sector. The intention of the report is to provide further insight into the potential effects on the New Zealand electricity grid that may occur with a dramatic increase in EVs, so that these may be planned for and mitigated. It is also inspired by the UK Department of Transport 2018 statistical report (Eyers 2018).
The data used has been provided by ‘Flip the Fleet’, a community organisation that hopes to increase uptake of electric vehicles in New Zealand. Flip the Fleet have been collecting data on electric vehicle usage patterns, via Exact IOT Limited’s blackbox recorder, a small electronic device that connects to the vehicle’s internal computer and sends detailed data about the battery health, power demand, charging rate, speed and other performance information to a secure database.
The subset of this data provided to the University of Otago was collected from 50 domestic electric vehicles monitored from April 2018 to January 2019. The data consisted of 1,515,812 1 minute interval observations of timestamped odometer readings (in km) together with measurements of charging power (kW) and battery charge state (% charged) linked to a unique anonymised vehicle identifier.
There are a number of important limitations to this data:
Even though the use of an anonymised vehicle identifier should prevent the identification of the vehicles in the sample, the fine-grained temporal nature of the data and the relatively small population of EV owners from whom the sample is drawn (Flip The Fleet members) means that the data cannot be publicly released.
Figure 2.1 shows the number of unique EVs observed by time of day and date. As we can see the early part of the sample is sparse and indeed the maximum number of EVs observed in any 15 minute time period was only 22 out of a possible total of 50. While this will not affect some analyses, it is likely to introduce error and small sample effects to summary analyses (e.g. means) or month by month analyses. In some sections the analysis will therefore be restricted to the data from September to January.
In addition Table 2.1 shows that a small number of EVs have very few observations, in some cases not extending beyond 1 day (shown as 0 days observed).
Figure 2.1: Number of unique EVs observed by time of day and date
| id | nObs | startTime | endTime | meanWhCharging | maxWhCharging | nDaysObserved |
|---|---|---|---|---|---|---|
| 0cc746a3f5ae75ee94068a8354b6be08 | 3 | 2018-09-09 10:46:30 | 2018-09-09 10:48:42 | 0.0000000 | 0.000000 | 0 days |
| 01583b8a5f0344cc4aa3b3939a27af2a | 4 | 2018-09-09 10:34:12 | 2018-09-09 10:36:25 | 0.0000000 | 0.000000 | 0 days |
| 4a6bb6e7ffc28d9d8eda7b4c6377a027 | 19 | 2018-09-08 08:48:38 | 2018-09-09 10:27:50 | 4.2251742 | 27.557201 | 1 days |
| 126c8759ec95ba40070b16a11fe0e587 | 258 | 2018-09-30 11:54:18 | 2018-09-30 19:24:05 | 1.5869526 | 1.960213 | 0 days |
| 4e48f4155c29c763ffe6d9e17a495200 | 530 | 2019-01-17 14:12:57 | 2019-01-25 10:31:16 | 0.0000000 | 0.000000 | 8 days |
| 6e3293c77f562262ed6608db1b596d36 | 4315 | 2018-05-15 14:48:15 | 2018-12-06 13:25:56 | 0.2872577 | 47.245786 | 205 days |
Further, as Table 2.1 also indicates, there were several (6) vehicles that had no recorded charging observations. These were discarded (which also discarded those with very few observations).
We then discarded:
This left 44 remaining vehicles, and 1,291,881 observations as shown in Table 2.2.
| Weekdays | Weekends | NA | Sum | |
|---|---|---|---|---|
| Fast charging | 5830 | 2537 | 0 | 8367 |
| Not charging | 402519 | 98971 | 0 | 501490 |
| Standard charging | 584821 | 197203 | 0 | 782024 |
| NA | 0 | 0 | 0 | 0 |
| Sum | 993170 | 298711 | 0 | 1291881 |
Charging data has been broadly separated into two separate categories, ‘Standard’ and ‘Fast’. Standard charging is defined to be when the charger is reading less than 7kW - this is considered the upper limit of ordinary home charging without an expensive wiring upgrade (Concept Consulting 2018). Fast charging is all charging equal to or greater than 7kW, and would likely occur at designated and purpose-built public charging stations.
It should be noted that this method is not always accurate since we can identify apparent sequences of charging which start at > 7kW and decline to < 7kW over a relatively short period. In this circumstance the first observation will be correctly classified as ‘Fast’ but the lower observations, which we assume are lower power trickle ‘top-up’ at the end of a fast charge will be incorrectly classified as ‘Standard’ (see 8.1.2). This is clarified in Section 5 where we use the first observation in a sequence to denote fast/standard but has yet to be resolved in other sections. As a result we may currently be under-estimating the number of fast charge observations and over-estimating the mean power demand of standard charges. Future work will resolve this potential misclassification error.
As an example, we know that there are 105 sequences of charging events (out of a total of 15186) where the first and last charge types do not match. Of these 478 were pairs where the first charging observation was ‘Fast’ and the last classified at ‘Standard’.
Figure 2.2 shows the distribution of observed charging kW demand by inferred charge type without correcting for potential mis-classifications. Setting aside the small number of potential misclassifications noted above, the plot confirms the validity of our definition and shows that fast charges were relatively rare in the dataset. Fast charges have two distinct power demand ‘peaks’ at ~22kW and ~45kW while the far more common standard charging was mostly concentrated around 1.8kW and 3kW, with a smaller concentration around 6kW.
Figure 2.2: Observed power demand distribution by charge type where charging observed
In order to determine charging durations, we need to identify observations which are the start and end of charging sequences. We use the following logic to do this:
| Fast charging | Not charging | Standard charging | NA | Sum | |
|---|---|---|---|---|---|
| Charging in a seq | 7272 | 0 | 762195 | 0 | 769467 |
| First charge obs in a seq | 478 | 0 | 7110 | 0 | 7588 |
| Last charge in a seq | 402 | 0 | 7196 | 0 | 7598 |
| Not charging (0 kW) | 0 | 501490 | 0 | 0 | 501490 |
| Single charge observation | 213 | 0 | 5513 | 0 | 5726 |
| NA | 2 | 0 | 10 | 0 | 12 |
| Sum | 8367 | 501490 | 782024 | 0 | 1291881 |
Table 2.3 shows the results of this coding for all clean observations. As we can see very few observations were not coded using this scheme. As shown in Section 8.1.2.1 an alternative method which added a 120 second maximim threshold to sequences of observations was also tested but not used as it failed to identify sparse sequences of charging events.
Using this method we obtained 7,588 instances of charging starting, and 7,598 instances of charge ending. The additional 10 instances of the charge ending than there are of the charge beginning may be due to the first instance of data collection occurring during mid-charge for some vehicles.
The charge duration was then calculated as being the time duration between each pair of “first charge” and “last charge” observations
Figure 2.3 shows the overall distribution of all charging sequences. Clearly there are very small and a few very large values for both charging types.
## Don't know how to automatically pick scale for object of type difftime. Defaulting to continuous.
## Warning: Removed 1 rows containing non-finite values (stat_bin).
Figure 2.3: Duration of charging sequences
Table 2.4 shows the overall distributions and indicates the extent to which the means are skewed by the very small and a few very large values shown in Figure 2.3.
| chargeTypeFixed | N | mean | median | min | max |
|---|---|---|---|---|---|
| Standard charging | 7095 | 98.49 mins | 3.43 | 0.27 mins | 1616.72 mins |
| Fast charging | 492 | 41.89 mins | 13.53 | 0.02 mins | 8621.00 mins |
Figure 2.4 shows the distribution of very short charging sequences. As we can see these appear to be generally less than 8 minutes in length for Standard Charges.
## Don't know how to automatically pick scale for object of type difftime. Defaulting to continuous.
Figure 2.4: Duration of charging sequences < 15 minutes
Table 2.5 shows the same descriptive statistics but for all sequences of greater than 8 minute duration. Now we can see that the mean and median durations for both Standard and Fast Charge sequences are closer.
| chargeTypeFixed | N | mean | median | min | max |
|---|---|---|---|---|---|
| Standard charging | 2809 | 245.40 mins | 209.80 | 8.02 mins | 1616.72 mins |
| Fast charging | 356 | 56.48 mins | 17.86 | 8.05 mins | 8621.00 mins |
Manual inspection of the data showed that these short-duration charging “events” generally occurred near the end of a longer-duration charging sequence It appeared that once the vehicle had reached its highest state of charge, charging would intermittently stop and start again. This is probably due to the behaviour of the charger once the battery was almost full. In addition to the myriad “short” charging duration values, a small number of unreasonably long charging durations (longer than 100 hours for standard charging or longer than 14 hours for fast charging) were calculated. As these exceeded the expected charge durations of the most high capacity vehicles currently available, they were also assumed to be anomalies. The analyses in Section 5 below was therefore made with the following charge events excluded from the data:
Figure 2.5 and 2.6 shows the distribution of charging sequences with the excessively long or short events removed. These charging durations appear more reasonable when considering standard battery capacities and charging powers.
## Don't know how to automatically pick scale for object of type difftime. Defaulting to continuous.
Figure 2.5: Duration of charging sequences with unreasonably long or short values removed
## Saving 7 x 5 in image
## Don't know how to automatically pick scale for object of type difftime. Defaulting to continuous.
| chargeTypeFixed | N | mean | median | min | max |
|---|---|---|---|---|---|
| Standard charging | 2810 | 245.32 mins | 209.73 | 8.00 mins | 1616.72 mins |
| Fast charging | 489 | 20.86 mins | 13.50 | 0.02 mins | 767.20 mins |
It has been suggested that EV charging is more likely to occur in the early evening when drivers return from daily commutes or school pick-ups (Langbroek, Franklin, and Susilo 2017). Figure 3.3 uses a density plot to represent the proportion of charging and non-charging observations at different times of day by weekday vs weekends. The plot clearly shows non-charging during day-time use and also shows a bi-model distribution for fast charging (non-corrected categorisation). Standard charging also shows a bi-modal distribution with a peak around 22:00 on weekdays and another at 01:00 presumably indicating the use of timed or ‘smart’ charging or trickle events.
Figure 3.1: Density plot of charging start times during weekdays
## Saving 7 x 5 in image
These patterns are also visible in Figure 3.1 which shows the distribution of observed charging events by time of day and day of the week.
Figure 3.2: Count of observed charging events by type, day of week and time
This figure indicates that the greatest frequency of standard charging events occur between the hours of 8pm and 8am, with very low occurrences of charging during morning and evening grid peaks. Fast charging on the other hand is a day-time activity on both weekdays and weekends.
To make the patterns of ‘initial charging’ clearer, we use just the ‘first’ charge observation in a pair (see above) and exclude automatic battery ‘top-ups’ (refer to Section 6) by also filtering out any data where a charging observation begins while the state of charge is greater than 90%. Having done so, Figure 3.3 uses a density plot to represent the proportion of charging events that begin at different times of the day on weekdays vs weekends for standard and fast charging (corrected classification).
Figure 3.3: Density plot of charging start times during weekdays where state of charge < 90%
## Saving 7 x 5 in image
As we can see, standard charging sequences (as opposed to single observations) have a noticeably different profile to charging patterns for fast charges. It suggests that the majority of standard charging events start at 22:00 and run overnight at home, and perhaps use the more powerful public charge points to top up during the day. However the plot also shows that this is not universal with a reasonable proportion of charging events starting earlier in the day, including during the NZ peak demand periods of 07:00 - 09:00 and 17:00 - 21:00.
Standard charging events were most likely to begin around 10pm during both weekdays and weekends. As it seems unlikely that this is due to vehicle drivers returning home at this hour, this effect may be due to drivers setting the charger on a timer to take advantage of cheaper “off-peak” electricity times, which frequently begin around 10pm.
Fast charging events were most likely to begin at 11:30am on weekdays and 1pm during weekends.
Given this distribution of charging events, it is important to understand their magnitude to understand the potential effect on the electricity network. Although we are hampered by the lack of ‘no charge’ data when the EV is not connected to the charger and switched off, this section analyses the patterns of power demand where charging is observed. Clearly this does not provide overall sample mean power demand which would include charging, non-charging and non-use observations.
Overall 75% of standard charging observations were 1.47 kW or more but the figure was 20.28 kW or more for fast charging.
Figure 4.1 shows the mean power demand for standard charging observations by time of day and weekdays vs weekends for the charging data collected after September 2018 to ensure maximum sample size (see Section 2.2). The plot uses transparency to indicate the number of EVs contributing to each of the mean calculations to give a guide to their reliability. Dots with stronger colours indicate means calculated from a larger number of EVs and, given the data gaps noted in Section 2.1, this indicates patterns which are generally shared across more EVs.
This plot appears to show that there are three peaks in standard charging, one at 10:00, one at 18:00 (possibly based on fewer EVs) and one after midnight on weekdays. There are also noticeable 07:00 and 16:00 charging blips. On the other hand at weekends the daytime peak shifts to 14:00. Thus, while our previous analysis suggested that charging events were more likely to start later in the evening, the power demand of earlier charging events may actually be relatively high and co-incide with exisitng peak demand periods.
Figure 4.1: Mean charging power demand (kW) by time of day
Fast charging however has no detectable pattern other than a clear increase in density during weekday daytimes (Figure 4.2).
Figure 4.2: Mean charging power demand (kW) by time of day
It is possible that the ‘standard charge’ day-time peak is skewed by mis-classified short low power ‘fast charge’ observations (see Section @ref(#chargeType)). Figure 4.3 attempts to allow for this misclassification by plotting the median rather than the mean. The plot more clearly shows the 10:00 weekday spike which, if we assume that the mis-classified ‘fast charges’ will be skewing the standard charge mean value upwards, is likely to be due to mis-classified ‘fast charging’. However the 18:00 peak persists as does the 14:00 weekend peak while overnight charging levels are relatively stable as we would expect from 4.1.
Figure 4.3: Median charging power demand (kW) by time of day
Figure 4.4 repeats the median power-based analysis for ‘Standard charging’ but shows the results by month. While the sample size is probably too small to draw robust conclusions there appear to be differences between months with December showing few discernable peaks and September and January showing much lower daytime weekday charging. In addition, weekdays and weekends are much more similar in November and December.
Figure 4.4: Median charging power demand (kW) by time of day
On face value the results suggest that EVs could be placing additional power demand on local and national networks during well-known periods of peak demand although this appears to vary by month for this small sample of EV owners.
Clearly this analysis should be revisited once the potential misclassification of ‘fast’ as ‘standard’ charging observations has been resolved and the ‘missing’ non-use (zero charging) observations have been imputed.
This section analyses the duration of observed charging events to understand when longer charging sequences are likely to occur. Table 5.1 shows the mean durations for all all charging events by event start time for standard charging durations greater than 8 minutes (see Section 2.3.2) and all fast charging events for observations collected after 01 September 2018.
| chargeTypeFixed | mean | median | min | max | sd |
|---|---|---|---|---|---|
| Standard charging | 251.15 mins | 217.78 mins | 8.00 mins | 1616.72 mins | 189.62 |
| Fast charging | 21.66 mins | 13.70 mins | 0.02 mins | 767.20 mins | 53.25 |
| qHour | chargeTypeFixed | weekdays | meanDuration | nEVs |
|---|---|---|---|---|
| 10:30:00 | Standard charging | Weekends | 684.57 mins | 2 |
| 04:45:00 | Standard charging | Weekdays | 596.40 mins | 1 |
| 21:00:00 | Fast charging | Weekdays | 582.53 mins | 1 |
| 21:00:00 | Standard charging | Weekends | 500.20 mins | 11 |
Figure 5.1 plots the mean duration by time of day and weekday vs weekend and charge type. As before we use transparency to indicate the number of unique EVs contributing to the mean values and we have removed a small number of very large duration outliers (mean duration > 540 minutes or 9 hours) which appears to be based on just 1 or 2 EVs (see Table @ref:(tab:makeDurationTimeMean)).
As we would expect, the plot shows that for standard charging mean ‘forward’ duration generally decreases from midnight, presumably as batteries are becoming fully charged through to 06:00 and then increases as the time of starting to charge increases through the day before trending downwards before midnight. Again, this confirms that charge events starting in or just after the evening peak demand period on both weekdays and weekends are likely to be longer, possibly reflecting the lower state of charge at this time of day (following use).
Duration of fast charge events by start time appear to be more randomly distributed, although very few events were recorded between midnight and 7am. This, along with the comparatively low number of recorded fast charge events indicated in Fig. 2.2 suggests that drivers utilize fast charging only “as necessary” to ensure they have enough battery capacity to complete their journey or when ‘at work’ or conducting some other mobility related task such as shopping.
## Don't know how to automatically pick scale for object of type difftime. Defaulting to continuous.
Figure 5.1: Mean duration (within quarter hours) by time of charging start
The state of charge is the percentage of energy still available to be used in the battery. In future, electric vehicles may be able to discharge any remaining battery charge as electricity into the grid, a process known as vehicle to grid (V2G) energy transfer. This may allow electric vehicles to have a net beneficial effect on the grid, reducing the evening peaks by providing electricity to the home during this period, and then recharging later in the evening or early the next morning when peak demand has diminished.
This section provides an indication of the state of charge of electric vehicles upon charging, so that the potential of V2G technology can be assessed.
Figure 6.1: Value of state of charge at beginning of charge
## Saving 7 x 5 in image
As can be seen in Figure 6.1, using the cleaned complete observations data, the majority of standard charges begin while the state of charge is above 90%. This is most likely due to the manner in which the charger regularly turns off and on again near the end of the charging cycle as described in Section 2.2.
Figure 6.2 shows the state of charge values when charge begins but with state of charge greater than 90% removed from the data for clarity. The figure indicates that many vehicles begin charging despite having greater than 50% charge remaining. This has clear implications for battery life management since continually top-up charging is known to substantially shorten the lifetime of EV batteries (XX ref needed XX). However it also indicates the potential to use the charge in the battery to feed into the grid, especially in the residential context.
Figure 6.2: Value of state of charge at beginning of charge (values > 90% removed)
## Saving 7 x 5 in image
Figure 6.3: Value of state of charge at beginning of charge (values > 90% removed)
## Saving 7 x 5 in image
Figure 6.3 repeats this analysis but uses the cleaned and corrected inferred start/end of charging sequence data instead of all charging observations. Figure 6.3 shows very similar distributions to the previous ‘all-observations’ plot (Figure 6.2) and confirms that sequences of standard charging in particular most frequently start with battery state of charge over 50%.
Figure 6.4: Mean state of charge at beginning of charge (values > 90% removed)
## Saving 7 x 5 in image
Finally, Figure 6.4 shows the mean % charge by time of first charging observation in a sequence using the cleaned and corrected inferred start/end of charging sequence data. The plot suggests that this capacity may be relatively stable throughout the day albiet with slightly higher mean capacity around the morning peak as we would expect given over-night charging. It is unlikely that this early morning capacity would be willingly made available for V2G since the EV may be used in the near future although this may not always be the case. However it is interesting to note that mean capacity at start of charge in the evening peak period is still roughly 50% indicating relatively substantial power availability.
Based on a relatively small and probably non-representative sample of 44 domestic electric vehicles provided by our research partner FlipTheFleet and which were monitored from April 2018 to January 2019 we have found that:
In the data provided for this study, most charging occurs at home using either a 1.8kw or 3kW charger, and commonly occurs both in the evening peak period and through the night. In addition, many vehicles begin charging with significant battery capacity remaining, providing them with the ability to provide vehicle to grid energy transfer should that technology become widely available.
These preliminary findings support recent modelling work (Concept Consulting 2018) that suggests that any negative effects electric vehicles may have on the evening national electricity grid peaks should be mitigable through ‘smart’ charging methods. In addition, our analysis indicates that this may already be occurring to some extent in this sample of EV owners. If later adopters of electric vehicles can be induced to follow the same ‘smart’ charging patterns as those displayed in some of our data sample, it is likely that the effects that electric vehicles are otherwise likely to have on the electricity grid may be mitigated.
Data description for original data supplied (before processing or filtering).
## Skim summary statistics
## n obs: 1515812
## n variables: 8
##
## ── Variable type:character ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max empty n_unique
## day_of_week 0 1515812 1515812 6 9 0 7
## id 0 1515812 1515812 32 32 0 50
## month 0 1515812 1515812 3 3 0 10
##
## ── Variable type:Date ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max median
## date 0 1515812 1515812 2018-04-05 2019-01-25 2018-11-09
## n_unique
## 293
##
## ── Variable type:difftime ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max median n_unique
## time 0 1515812 1515812 0 secs 86399 secs 44827 secs 86400
##
## ── Variable type:numeric ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n mean sd p0
## charge_power_kw 0 1515812 1515812 1.73 71 0
## odometer_km 1000156 515656 1515812 7290.5 7954.38 -62920
## state_of_charge_percent 0 1515812 1515812 69.11 20.85 0
## p25 p50 p75 p100 hist
## 0 1.37 1.9 74940.42 ▇▁▁▁▁▁▁▁
## 1889 4749 10529 69394 ▁▁▁▆▇▂▁▁
## 56.43 70.57 83.2 1677.72 ▇▁▁▁▁▁▁▁
Data description for cleaned data (all observations).
## Skim summary statistics
## n obs: 1291881
## n variables: 15
##
## ── Variable type:character ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max empty n_unique
## chargeFlag 12 1291869 1291881 17 25 0 5
## chargeType 0 1291881 1291881 12 17 0 3
## dvID 0 1291881 1291881 9 10 0 44
## id 0 1291881 1291881 32 32 0 44
## weekdays 0 1291881 1291881 8 8 0 2
##
## ── Variable type:Date ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max median
## date 0 1291881 1291881 2018-05-12 2019-01-25 2018-11-13
## n_unique
## 249
##
## ── Variable type:difftime ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max median
## dateTimeDiff 44 1291837 1291881 0 secs 4912664 secs 50 secs
## qHour 0 1291881 1291881 0 secs 85500 secs 11:45:00
## time 0 1291881 1291881 0 secs 86399 secs 11:50:53
## n_unique
## 13443
## 96
## 86400
##
## ── Variable type:factor ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n n_unique
## day_of_week 0 1291881 1291881 7
## month 0 1291881 1291881 9
## top_counts ordered
## Fri: 206422, Wed: 206292, Thu: 205166, Mon: 190389 TRUE
## Nov: 298332, Oct: 280869, Dec: 272272, Jan: 188498 FALSE
##
## ── Variable type:numeric ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n mean sd p0 p25
## charge_power_kw 0 1291881 1291881 1.48 2.89 0 0
## odometer_km 857152 434729 1291881 6801.77 7943.05 -62920 1698
## SoC_percent 46 1291835 1291881 68.42 18.58 0 55.68
## p50 p75 p100 hist
## 1.38 1.9 70.16 ▇▁▁▁▁▁▁▁
## 4123 8816 69394 ▁▁▁▇▇▂▁▁
## 69.73 82.26 98.1 ▁▁▂▃▆▇▇▇
##
## ── Variable type:POSIXct ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max median
## dateTime 0 1291881 1291881 2018-05-12 2019-01-25 2018-11-13
## n_unique
## 1230977
Data description for cleaned data (first observations in a charging sequence).
## Skim summary statistics
## n obs: 3299
## n variables: 19
##
## ── Variable type:character ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max empty n_unique
## chargeFlag 0 3299 3299 25 25 0 1
## chargeType 0 3299 3299 13 17 0 2
## chargeTypeError 0 3299 3299 29 37 0 4
## dvID 0 3299 3299 9 10 0 43
## endType 0 3299 3299 13 17 0 2
## id 0 3299 3299 32 32 0 43
## weekdays 0 3299 3299 8 8 0 2
##
## ── Variable type:Date ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max median n_unique
## date 0 3299 3299 2018-05-12 2019-01-17 2018-11-09 201
##
## ── Variable type:difftime ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min
## dateTimeDiff 0 3299 3299 0 secs
## pairDuration 0 3299 3299 0.01666667 mins
## qHour 0 3299 3299 0 secs
## time 0 3299 3299 40 secs
## max median n_unique
## 230025 secs 305 secs 1804
## 1616.717 mins 178.8167 mins 3024
## 85500 secs 15:15:00 96
## 86246 secs 15:22:14 3167
##
## ── Variable type:factor ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n n_unique
## chargeTypeFixed 0 3299 3299 2
## day_of_week 0 3299 3299 7
## month 0 3299 3299 9
## top_counts ordered
## Sta: 2810, Fas: 489, Not: 0, NA: 0 FALSE
## Fri: 524, Wed: 502, Thu: 490, Mon: 489 TRUE
## Nov: 791, Oct: 743, Dec: 702, Sep: 458 FALSE
##
## ── Variable type:numeric ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n mean sd p0 p25
## charge_power_kw 0 3299 3299 7.31 12.68 0.5 1.62
## odometer_km 2555 744 3299 5649.15 7459.93 -52352 1289.25
## SoC_percent 0 3299 3299 49.41 18.72 4.11 35.89
## p50 p75 p100 hist
## 2.62 3.35 70.16 ▇▁▁▁▁▁▁▁
## 3435 7345.75 54443 ▁▁▁▂▇▁▁▁
## 48.41 59.63 98.1 ▁▃▆▇▇▃▂▂
##
## ── Variable type:POSIXct ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────
## variable missing complete n min max median n_unique
## dateTime 0 3299 3299 2018-05-12 2019-01-17 2018-11-09 3299
This is used to identify observations that form part of a sequence. The logic is given in Section 2.3.2. Here we show the results of applying an additional 120 second rule. In this case a sequence only exists where we have charging observations which have less than 120 seconds between them.
| Fast charging | Not charging | Standard charging | NA | |
|---|---|---|---|---|
| Charging in a seq | 6995 | 0 | 750485 | 0 |
| First charge obs in a seq | 311 | 0 | 5124 | 0 |
| Last charge in a seq | 380 | 0 | 7045 | 0 |
| Not charging (0 kW) | 0 | 501490 | 0 | 0 |
| Not classified (what is this??) | 464 | 0 | 13839 | 0 |
| Single charge observation | 213 | 0 | 5513 | 0 |
| NA | 4 | 0 | 18 | 0 |
| Fast charging | Not charging | Standard charging | NA | |
|---|---|---|---|---|
| Charging in a seq | 7272 | 0 | 762195 | 0 |
| First charge obs in a seq | 478 | 0 | 7110 | 0 |
| Last charge in a seq | 402 | 0 | 7196 | 0 |
| Not charging (0 kW) | 0 | 501490 | 0 | 0 |
| Single charge observation | 213 | 0 | 5513 | 0 |
| NA | 2 | 0 | 10 | 0 |
As we can see, applying the 120 second rule reduces the number of observations categorised as part of a sequence as it will not know what to do with:
For now we therefore do not use the 120 second rule.
chargeType is used to classify charging events into standard vs fast using the 7 kW threshold. But there may be misclassfications where a sequence starts on a fast charger but power demand declines below the threshold. We can check this.
## chargeType is used to classify charging events into standard vs fast using the 7 kW threshold. But there may be misclassfications:
##
## Error: first = Fast, last = Standard
## Fast charging 91
## Standard charging 0
## <NA> 0
##
## Error: first = Standard, last = Fast
## Fast charging 0
## Standard charging 14
## <NA> 0
##
## OK: first = Fast, last = Fast
## Fast charging 387
## Standard charging 0
## <NA> 0
##
## OK: first = Standard, last = Standard <NA>
## Fast charging 0 402
## Standard charging 7096 7196
## <NA> 0 0
## There are 105 pairs (out of a total of 7593) from 26 EVs where charge type doesn't match.
## N observations where previous dateTime unknown (should match to n EVs)
## [1] 43
## N observations where next dateTime unknown (should match to n EVs)
## [1] 43
## N observations where chargeFlag is unknown
## [1] 9
## These seem to occur when charging is detected but the dateTime before/after is unkown due to data truncation
Check charge flags:
## chargeFlag is used to classify charging events - check against charge type:
##
## Standard charging <NA>
## Charging in a seq 734213 0
## First charge obs in a seq 6598 0
## Last charge in a seq 6678 0
## Single charge observation 4593 0
## <NA> 9 0
## There are a few observations that have chargeFlag = NA but are charging... why?
## N observations where previous dateTime unknown (should match to n EVs)
## [1] 43
## N observations where next dateTime unknown (should match to n EVs)
## [1] 43
## N observations where chargeFlag is unknown
## [1] 9
## These seem to occur when charging is detected but the dateTime before/after is unkown due to data truncation
Azadfar, Elham, Victor Sreeram, and David Harries. 2015. “The investigation of the major factors influencing plug-in electric vehicle driving patterns and charging behaviour.” Renewable and Sustainable Energy Reviews 42. Elsevier: 1065–76. doi:10.1016/j.rser.2014.10.058.
Concept Consulting. 2018. “‘ Driving change ’ – Issues and options to maximise the opportunities from large-scale electric vehicle uptake in New Zealand,” no. March.
Eyers, Lisa. 2018. “Electric Chargepoint Analysis 2017 : Domestics Key findings :” no. December.
Khan, Imran, Michael W. Jack, and Janet Stephenson. 2018. “Analysis of greenhouse gas emissions in electricity systems using time-varying carbon intensity.” Journal of Cleaner Production 184. Elsevier Ltd: 1091–1101. doi:10.1016/j.jclepro.2018.02.309.
Langbroek, Joram H. M., Joel P. Franklin, and Yusak O. Susilo. 2017. “When Do You Charge Your Electric Vehicle? A Stated Adaptation Approach.” Energy Policy 108 (September): 565–73. doi:10.1016/j.enpol.2017.06.023.
Li, Wenbo, Ruyin Long, Hong Chen, and Jichao Geng. 2017. “A Review of Factors Influencing Consumer Intentions to Adopt Battery Electric Vehicles.” Renewable and Sustainable Energy Reviews 78 (October): 318–28. doi:10.1016/j.rser.2017.04.076.
Rezvani, Zeinab, Johan Jansson, and Jan Bodin. 2015. “Advances in Consumer Electric Vehicle Adoption Research: A Review and Research Agenda.” Transportation Research Part D: Transport and Environment 34 (January): 122–36. doi:10.1016/j.trd.2014.10.010.
Speidel, Stuart, and Thomas Bräunl. 2014. “Driving and Charging Patterns of Electric Vehicles for Energy Usage.” Renewable and Sustainable Energy Reviews 40 (December): 97–110. doi:10.1016/j.rser.2014.07.177.
Stephenson, Janet, Rebecca Ford, Nirmal-Kumar Nair, Neville Watson, Alan Wood, and Allan Miller. 2017. “Smart grid research in New Zealand – A review from the GREEN Grid research programme.” doi:10.1016/j.rser.2017.07.010.
Transpower New Zealand. 2015. “Transmission Planning Report,” no. July: 320.
———. 2017. “Battery Storage in New Zealand,” no. September: 41. https://www.transpower.co.nz/sites/default/files/publications/resources/Battery Storage in New Zealand.pdf.